Different results for Static General, Dynamic Implicit and Dynami Explicit

[Zaurbek]

Dear Colleagues,
I'm doing a cyclic loading of RC column using Static General, Dynamic Implicit and Dynamic Explicit solvers.
Though I have almost similar results for the SG and DI cases, the results are very different for the DE case (see Figure). I've tried to divide each cycle into 4 load steps, but this also gives different results.
Then I used Smooth amplitude instead of Tabular and changed to Double Precision, but still got different results.
Can anyone suggest how to solve this problem?
Many thanks in advance!
I attach the following input files to Test-1.zip:
- Job-1: Cyclic loading with Static General solver
- Job-2: Cyclic loading with Dynamic Implicit solver
- Job-3: Cyclic loading with Dynamic Explicit solver (full smooth protocol)
- Job-4: Cyclic loading with Dynamic Explicit solver (divided smooth protocol)

 

[FEA way]

Interesting comparison. Abaqus/Explicit is a very different solver. It's inherently dynamic and you have to be really careful to avoid the influence of inertia forces if you are trying to run quasi-static analysis. Did you check all the energies in Abaqus/Explicit results ? This is crucial. Not only to make sure kinetic energy is small enough but also to verify the influence of automatically applied numerical damping among others.

 

[Zaurbek]

Interesting comparison. Abaqus/Explicit is a very different solver. It's inherently dynamic and you have to be really careful to avoid the influence of inertia forces if you are trying to run quasi-static analysis. Did you check all the energies in Abaqus/Explicit results ? This is crucial. Not only to make sure kinetic energy is small enough but also to verify the influence of automatically applied numerical damping among others.

Thanks for your questions.
I used a constant loading rate of 5 mm/sec when generating the cyclic amplitude to avoid inertia effects.
I noticed that all my explicit models have similar results with SG and DI steps up to 12 seconds,
and after that there is some distorted response and the curve degrades (see Figure 1).
This is strange because I use the same amplitude for all cases.
I checked the energies and found less than 5% of the kinetic energy demand (see Figure 2).
I then tried to refine the cyclic amplitude increments, but still got the same results.
The models give almost the same results for displacement up to 25 mm (see Figure 3),
but why do I get different results for quasi-static analysis?

 

[SWComposites]

So what is the issue? The results all look very close. Why would you expect them to be exactly the same? FEA is an approximate numerical solution to an approximate representation of reality.

 

[FEA way]

I checked the energies and found less than 5% of the kinetic energy demand (see Figure 2).

Kinetic energy looks good but you should check other energies as well. ALLAE is suspiciously high, there might be some hourglassing in your model. Also, check ALLVD (to make sure numerical damping is not too large, as I've mentioned above) and other relevant energies. This is really important in the case of Abaqus/Explicit analyses.

 

[Zaurbek]

Kinetic energy looks good but you should check other energies as well. ALLAE is suspiciously high, there might be some hourglassing in your model. Also, check ALLVD (to make sure numerical damping is not too large, as I've mentioned above) and other relevant energies. This is really important in the case of Abaqus/Explicit analyses.

Thank you for your advice. Hourglassing it was.
I changed the control from viscous to combined (with stiffness-viscous factor 0.5) and the response became stable.

 

[FEA way]

Good to know what the problem was. Once again, checking all the relevant energies in Explicit is a must.